Current slope risk assessment methods suffer from deficiencies in computational complexity and comprehensiveness, especially when applied to transmission tower foundation slopes. To address this issue, this study constructs a risk assessment index system for transmission tower foundation slopes from both hazard and health perspectives and proposes a risk assessment method based on residual neural networks and Bayesian optimization. The research results show that this method can achieve an accuracy of 98.4% in practical applications, significantly outperforming traditional BP neural networks and deep neural networks, and demonstrating strong generalization capabilities. Furthermore, the model was validated through recent rainfall-induced landslide events, further proving the method’s practicality and reliability. These results indicate that the Bayesian optimized residual neural network has significant advantages in handling complex slope risk assessment problems. It can be effectively applied to the risk asse